CN-115186576-B - Non-motor vehicle track prediction method

CN115186576BCN 115186576 BCN115186576 BCN 115186576BCN-115186576-B

Abstract

The invention relates to a non-motor vehicle track prediction method, which comprises the steps of calculating a feasible region of a non-motor vehicle end point to be detected based on scene information to be predicted, obtaining a feasible end point set through discrete sampling, generating an alternative track set of the non-motor vehicle to be detected according to the scene information to be predicted and the feasible end point set by adopting a pre-built and trained deep learning model, evaluating the alternative track set by adopting a pre-built behavior analysis model of the non-motor vehicle, and selecting a maximum utility track, wherein the behavior analysis model is used for calculating track utility which is quantized into heterogeneous summation of expected track risk and track efficiency. Compared with the prior art, the method can adapt to flexible and changeable behavior characteristics of the non-motor vehicle, has certain advantages in the aspects of eliminating accumulated errors in iterative prediction and track prediction aiming at abrupt behavior of the non-motor vehicle, and greatly improves the accuracy and reliability of the track prediction of the non-motor vehicle.

Inventors

NI YING
LI JIANQIANG
SUN JIAN

Assignees

同济大学
同济大学

Dates

Publication Date: 20260421
Application Date: 20220617
Priority Date: 20220617

Claims (7)

1. A method for predicting a trajectory of a non-motor vehicle, comprising the steps of: s1, calculating a feasible region of an end point of a non-motor vehicle to be detected based on scene information to be predicted, and obtaining a feasible end point set through discrete sampling; S2, adopting a pre-constructed and trained deep learning model, and generating an alternative track set of the non-motor vehicle to be tested according to scene information to be predicted and a feasible terminal set; S3, evaluating an alternative track set by adopting a pre-built behavior analysis model of the non-motor vehicle, and selecting a maximum utility track, wherein the behavior analysis model is used for calculating track utility which is quantized into heterogeneous summation of expected track risk and track efficiency; the calculation expression of the track utility is as follows: In the formula, Is the utility value of the ith trace, A coordinate matrix representing the ith track, R and L respectively represent path risk and path efficiency, wherein R is a negative value and represents the opposite direction to L; representing a riding preference factor representing individual heterogeneous pursuits for risk and efficiency; The calculation expression of the path risk is as follows: wherein R is the path risk, d is the path length; The risk value of the position of the jth track point is represented, and n track points are shared; E represents the risk value of a certain point, and is a scalar; And Respectively representing potential energy fields and dynamic fields, wherein the potential energy fields and the dynamic fields respectively represent risk distribution situations of static and dynamic objects as risk sources; The power field Based on the sequence prediction position calculation of the dynamic interaction object, the calculation expression of the sequence prediction position of the dynamic interaction object is as follows: In the formula, Representing the position matrix of each step length of the dynamic interactive object in the prediction time; Indicating its initial velocity; representing a time step matrix; Representing initial position coordinates; The calculation expression of the path efficiency is as follows: wherein L is path efficiency; For the forward displacement of the j-th step, Points are total track points.
2. The method for predicting the track of the non-motor vehicle according to claim 1, wherein the scene information to be predicted comprises non-motor vehicle information to be detected, other traffic participant information and environment information, and the predicting process of all feasible endpoints of the non-motor vehicle comprises: And removing the occupied areas of other traffic participants in the feasible area to obtain the feasible areas of all the endpoints of the non-motor vehicle to be detected.
3. The method for predicting the trajectory of a non-motor vehicle according to claim 2, wherein the calculation expression of the maximum and minimum displacement of the non-motor vehicle to be measured in the advancing direction and the transverse direction is: In the formula, 、、 And The furthest and closest distances the non-motor vehicle to be tested can reach in the predicted time along the x axis and the y axis respectively; And The current x and y coordinates of the non-motor vehicle to be tested are respectively; And The current speeds of the non-motor vehicle to be measured in the x-axis direction and the y-axis direction are respectively; 、、 And Maximum acceleration of the non-motor vehicle to be tested in positive and negative directions of the x axis and the y axis respectively; the calculation expression for removing the occupation area of other traffic participants in the feasible domain is as follows: In the formula, And Representing the position of an object in a feasible region of the non-motor vehicle to be tested; Expressed in terms of A spatial shape that is a centroid; Being an empty set, indicates that the region does not fit within the feasible region.
4. The method for predicting the track of the non-motor vehicle according to claim 1, wherein the discrete sampling is specifically that space discrete sampling is performed based on a diamond-shaped occupation, the diamond-shaped occupation is dynamically variable, and the expression of the diamond-shaped occupation is as follows: In the formula, Is the face of the sampling space; Representing diamond-shaped areas, wherein The long axis of the diamond is used for describing the sampling of the space of the advancing direction of the non-motor vehicle to be tested; A short axis of the diamond shape describes the sampling of the transverse space of the movement of the non-motor vehicle to be measured, wherein By adjusting And The value of (2) changes the concentration of spatial sampling, and realizes the balance between the running speed and the prediction accuracy of the algorithm.
5. The method for predicting the track of the non-motor vehicle according to claim 1, wherein the deep learning model is used for generating midpoint information of the track according to input track starting point information and track end point information, and the generation process of the alternative track set of the non-motor vehicle to be detected specifically comprises the following steps: According to a starting point and a feasible end point set of the non-motor vehicle to be detected, adopting a track generation algorithm, and generating a track end point through continuous iteration of the deep learning model so as to obtain a non-motor vehicle track; and traversing each endpoint in the feasible endpoint set to obtain the alternative track set.
6. The method of claim 5, wherein the calculation of the trajectory generation algorithm comprises the steps of: s201, selecting an ith endpoint in a feasible endpoint set as a current track generation endpoint, and setting an iteration algebra K; S202, generating a complete starting and ending point characteristic vector based on starting point information and known ending point positions; s203, loading the starting and ending point feature vector into a trained deep learning model to generate absolute position coordinates of the middle track point; s204, converting the absolute position coordinates of the intermediate track points into proportional coordinates; S205, selecting an end point position according to all the known track point information, returning to the step S202, and executing ; S206, judging whether k=0 is satisfied, if so, executing step S207, and if not, executing step S205 in a loop; s207 setting up Steps S201 to S206 are repeatedly performed until all the generation tasks of the preset trajectory set are completed.
7. The method of claim 5, wherein the complete start-end feature vector includes parameters of start and end positions of the non-motor vehicle and the interactive object to be tested, the parameters including coordinates, speed, acceleration, curvature and vehicle type; the calculation expression of the speed and the acceleration of the end positions of the non-motor vehicle to be measured and the interactive object is as follows: Wherein P, V, A, s respectively represents position coordinates, speed, acceleration and vehicle type, wherein P= { x, y }, V= { v_x, v_y }, A= { a_x, a_y }, respectively represent the position, speed and acceleration of a research object on x and y coordinate axes, s= {0,1,2}, respectively represent motor vehicles, electric bicycles and conventional bicycles, superscript host and int respectively represent research subjects and interactive objects, subscripts start, end and mid respectively represent starting and ending points and intermediate track points, t represents prediction time, and the interactive object is one other traffic participant closest to the Euclidean distance of a non-motor vehicle to be detected at the current moment.

Description

Non-motor vehicle track prediction method Technical Field The invention relates to the field of traffic flow track prediction, in particular to a non-motor vehicle track prediction method. Background While the automatic driving has entered the open road test stage, the open road has a large amount of non-shared space, and the characteristics of undefined driving road weights, no lane rules and the like cause the shared space such as intersections and the like to have complex interaction behaviors of various traffic participants, which becomes the challenge of the automatic driving vehicle to drive the open road. In view of the above challenges, predicting the behavior of other traffic participants is an effective way for an autonomous vehicle to achieve safe interaction and avoid collision accidents within a shared space. However, the understanding of complex situations and the processing ability of emergency situations displayed by the automated driving test reports are still the urgent problem to be solved for the automated driving vehicles, wherein the characteristics of non-motor vehicles such as strong maneuverability, random riding, abrupt behavior and the like are one of the main difficulties of track prediction Currently, research on non-motor vehicle trajectory prediction can be divided into two categories, including prediction methods based on physical models and prediction methods based on data driving. The physical model comprises a kinematic model and a dynamic model, wherein the main prediction principle is to extrapolate trend based on a historical motion state, the data-driven prediction method is mainly based on deep learning, training learning is performed by taking a historical track and a future track as input and output, and iterative prediction is performed. However, the current iterative prediction mode focuses more on deduction of the historical situation, but unavoidable accumulated errors exist in the iterative prediction process, so that the method is more suitable for behaviors with strong regularity, and the method is not suitable for the behavior of a motor vehicle, which is caused by the fact that the track has large difference and even abrupt change due to flexible and changeable behaviors. Therefore, the current track prediction method cannot fully meet the prediction requirements of the non-motor vehicle. Disclosure of Invention The invention aims to provide a non-motor vehicle track prediction method for overcoming the defects of accumulated errors caused by iterative prediction and abrupt behavior defects that a non-motor vehicle cannot be effectively predicted in the existing non-motor vehicle track prediction technology. The aim of the invention can be achieved by the following technical scheme: A non-motor vehicle track prediction method is suitable for mixed traffic conditions of intersections, and comprises the following steps: S1, obtaining scene information to be predicted, calculating a feasible region of a non-motor vehicle end point to be detected, and obtaining a feasible end point set through discrete sampling; S2, adopting a pre-constructed and trained deep learning model, and generating an alternative track set of the non-motor vehicle to be tested according to scene information to be predicted and a feasible terminal set; and S3, evaluating the alternative track set by adopting a pre-constructed behavior analysis model of the non-motor vehicle, and selecting the maximum utility track, wherein the behavior analysis model is used for calculating track utility which is quantized into heterogeneous summation of expected track risk and track efficiency. Further, the scene information to be predicted comprises information of the non-motor vehicle to be predicted, information of other traffic participants and environmental information, and the prediction process of all feasible endpoints of the non-motor vehicle comprises the following steps: And removing the occupied areas of other traffic participants in the feasible area to obtain the feasible area of the end point of the non-motor vehicle to be detected. Further, the calculation expression of the maximum and minimum displacement of the non-motor vehicle to be tested in the advancing direction and the transverse direction is as follows: Wherein x max、xmin、ymax and y min are the furthest and closest distances that the non-motor vehicle to be tested may arrive in the predicted time along the x-axis and the y-axis, respectively, x current and y current are the current x and y coordinates of the non-motor vehicle to be tested, respectively; And The current speeds of the non-motor vehicle to be measured in the x-axis direction and the y-axis direction are respectively; And The method comprises the following steps of respectively measuring the maximum acceleration of the non-motor vehicle to be measured in the positive direction and the negative direction of the x axis and the y axis, wherein the calculation express